DSANet: Dual-Branch Shape-Aware Network for Echocardiography Segmentation in Apical Views

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Accurate segmentation of cardiac anatomical structures in echocardiography is crucial for diagnosing heart diseases and assessing cardiac function.
• Challenges in echocardiography segmentation include unclear boundaries and significant shape deformations caused by cardiac motion, hindering automated analysis.

Purpose of the Study:

• To propose a novel Dual-Branch Shape-aware Network (DSANet) for automated segmentation of the left ventricle, left atrium, and myocardium in echocardiographic images.
• To enhance feature representation and segmentation performance by integrating shape-aware modules, an anisotropic strip attention mechanism, and cross-branch skip connections.

Main Methods:

• The proposed DSANet features a dual-branch architecture designed to capture shape priors and anatomical dependencies.
• An anisotropic strip attention mechanism and cross-branch skip connections are employed to improve feature integration and representation.
• A boundary-aware rectification module and boundary loss function were developed to enhance segmentation accuracy at ambiguous boundaries.

Main Results:

• DSANet demonstrated superior performance in segmenting cardiac structures compared to existing state-of-the-art methods.
• Experimental evaluations on public and in-house echocardiography datasets validated the effectiveness of the proposed network.
• The method showed significant improvements in accurately delineating cardiac anatomical structures.

Conclusions:

• The developed DSANet effectively addresses the challenges of cardiac segmentation in echocardiography.
• The network's ability to leverage shape information and refine boundary predictions offers a promising advancement for automated echocardiographic analysis.
• DSANet shows significant potential for improving the diagnosis and monitoring of cardiac diseases through enhanced segmentation accuracy.